Flow

What's Flow

Flow is an utility/ design pattern that help developers to write simple and readable code. There are two main concerns: Flow of operations and Flow of data

By using Flow, we should able to achieve the followings:

• Logics / operations can be reused easily
• The logic flows are readable by anyone (including the code reviewers)
• Each line of code is meaningfully and avoid ambiguous keywords
• No more callback hell for complicated async operations
• Debuggable both at development and production stage

Flow is referencing Composite pattern (https://en.wikipedia.org/wiki/Composite_pattern) and Chain-of-responsibility pattern (which including Command pattern) (https://en.wikipedia.org/wiki/Chain-of-responsibility_pattern)

So, we encapsulate operations as objects which can be chained using tree structures. Each operation is independent but able to be used with another one if the data required by the operations exist.

Here is an example for simple usage:

@IBAction func simpleChainedFlow() {
    Flow()
      .add(operation: SimplePrintOp(message: "hello world"))
      .add(operation: SimplePrintOp(message: "good bye"))
      .setWillStartBlock(block: commonWillStartBlock())
      .setDidFinishBlock(block: commonDidFinishBlock())
      .start()
  }

In these 5 lines of code, we can know that two operations will be executed in serial and able to do something before and after the operations.

Naming

To make the logic readable, it is important to make the operation's name meaningfully. It is developer's responsibility to make a good name. Also, the name also determine the degree of reusable of code. e.g. If you create an operation named: SimplePrintOp, it should contain only the code to print the message associated with it. You should NOT do anything out of the context of the name. Such as sending the message to server / write to file.

Also, all operations made for Flow should share a common suffix (e.g. Op) so all developers can know that there are operations that ready for reuse.

Grouped Operations

You can run a batch of operations using FlowArrayGroupDispatcher.

Flow()
      .setDataBucket(dataBucket: ["images": ["a", "b", "c", "d"]])
      .addGrouped(operationCreator: UploadSingleImageOp.self, dispatcher: FlowArrayGroupDispatcher(inputKey: "images", outputKey: "imageURLs", maxConcurrentOperationCount: 3, allowFailure: false))
      .start()

FlowArrayGroupDispatcher will dispatcher the targeted array in the data bucket to created operations and pass them the required data and collect them afterwards.

import Flow_iOS
class UploadSingleImageOp: FlowOperation, FlowOperationCreator {
  
  static func create() -> FlowOperation {
    return UploadSingleImageOp()
  }
  
  override var primaryInputParamKey: String { return "targetImageForUpload" }
  override var primaryOutputParamKey: String { return "uploadedImageUrl" }
  
  override func mainLogic() {
    guard let image: String = getData(name: primaryInputParamKey) else { return }
    DispatchQueue.global().asyncAfter(deadline: .now() + .seconds(1)) {
      self.log(message: "simulation of upload single image callback")
      self.setData(name: self.primaryOutputParamKey, value: "url_of_" + image)
      self.finishSuccessfully()
    }
    startWithAsynchronous()
  }
}

For the above example, FlowArrayGroupDispatcher will create a group of UploadSingleImageOp based on the array size of images in data bucket. As UploadSingleImageOp declares targetImageForUpload as it's input key and uploadedImageUrl as it's output key. FlowArrayGroupDispatcher will create temporary data bucket for each UploadSingleImageOp and contains targetImageForUpload inside. If the operation is succeed, FlowArrayGroupDispatcher will collect the object keyed with uploadedImageUrl and put into the result array imageURLs.

In such design, UploadSingleImageOp can be reused as single operation or grouped operation.

You can also set the maxConcurrentOperationCount (optional, default = 3) to control whether the operations are executed on one by one or in batch. If allowFailure (optional, default = false) is set to true, the Flow will continue to run even some / all operations in the group are failed. Therefore, the output array may be shorter than the input array or even empty.

Cases

Flow allow simple cases handling. For example:

@IBAction func demoCases() {
    Flow()
      .setDataBucket(dataBucket: ["images": ["a", "b", "c", "d", 1], "target": "A"])
      .add(operation: SimplePrintOp(message: "Step1"))
      .add(operation: SimplePrintOp(message: "Step2A1"), flowCase: FlowCase(key: "target", value: "A"))
      .add(operation: SimplePrintOp(message: "Step2A2"))
      .add(operation: SimplePrintOp(message: "Step2B1"), flowCase: FlowCase(key: "target", value: "B"))
      .add(operation: SimplePrintOp(message: "Step2B2"))
      .combine()
      .add(operation: SimplePrintOp(message: "Step3"))
      .setWillStartBlock(block: commonWillStartBlock())
      .setDidFinishBlock(block: commonDidFinishBlock())
      .start()
  }

After Step1 is finished, the Flow will run Step2A1 branch or Step2B1 branch depend on the value of target in data bucket. And combine is used to combine all cases back to a single node Step3.

To make the blueprint readable, nested case is NOT supported. Also, the type of case value must be String.

Data Handling

import Flow_iOS

class MockAsyncLoginOp: FlowOperation {
  override func mainLogic() {
    guard let email: String = getData(name: "email") else { return }
    guard let password: String = getData(name: "password") else { return }
    
    DispatchQueue.global().asyncAfter(deadline: .now() + .seconds(1)) {
      self.log(message: "simulation of login callback")
      if email == "[email protected]" && password == "123456" {
        let mockAccessToken = "sdaftadagasg"
        self.setData(name: "accessToken", value: mockAccessToken)
        self.finishSuccessfully()
      } else {
        let error = NSError(domain: "No such account", code: 404, userInfo: nil)
        self.finishWithError(error: error)
      }
    }
    startWithAsynchronous()
  }
}

In MockAsyncLoginOp in the example, it require two input data from data bucket (email and password). The Flow will check if the data exist and if the data type is correct (i.e. they must be String for this case). If no data is found with correct type, the operation is marked as failure and the Flow will stop. You can request any type you want. For example, you have a class named "LoginData" in your project.

guard let loginData: LoginData = getData(name: "loginData") else { return }

Making your Operations

Making operation is easy:

1. Pick a good name
2. Inherit FlowOperation
3. Put your logic inside mainLogic
4. For synchronized operation: call finishSuccessfully or finishWithError based on the result
5. For asynchronized operation: call startWithAsynchronous at the end of mainLogic after starting your async call
6. use log to record your debug logs
7. extends FlowOperationCreator to make the operation to use in Group And override primaryInputParamKey and primaryOutputParamKey

Some examples:

class SimplePrintOp: FlowOperation {
  var message: String!
  
  init(message: String) {
    self.message = message
  }
  
  override func mainLogic() {
    log(message: message)
    finishSuccessfully()
  }
}

class MockAsyncLoadProfileOp: FlowOperation {
  override func mainLogic() {
    guard let accessToken: String = getData(name: "accessToken") else { return }
    
    DispatchQueue.global().asyncAfter(deadline: .now() + .seconds(1)) {
      self.log(message: "simulation of success load profile callback")
      self.setData(name: "profileRefreshDate", value: Date())
      self.finishSuccessfully()
    }
    startWithAsynchronous()
  }
}

class UploadSingleImageOp: FlowOperation, FlowOperationCreator {
  
  static func create() -> FlowOperation {
    return UploadSingleImageOp()
  }
  
  override var primaryInputParamKey: String { return "targetImageForUpload" }
  override var primaryOutputParamKey: String { return "uploadedImageUrl" }
  
  override func mainLogic() {
    guard let image: String = getData(name: primaryInputParamKey) else { return }
    DispatchQueue.global().asyncAfter(deadline: .now() + .seconds(1)) {
      self.log(message: "simulation of upload single image callback")
      self.setData(name: self.primaryOutputParamKey, value: "url_of_" + image)
      self.finishSuccessfully()
    }
    startWithAsynchronous()
  }
}

Callbacks

You can set WillStartBlock and DidFinishBlock to get notified before or after the Flow run. They are called in main thread so you can do your UI changes. With the Flow instance in the block, you can get dataBucket: [String: Any], isSuccess: Bool and error: Error? and do your handling.

It's recommended to make common handling blocks which can further simplify your blueprint.

  func commonWillStartBlock(block: FlowWillStartBlock? = nil) -> FlowWillStartBlock {
    let result: FlowWillStartBlock = {
      flow in
      block?(flow)
      self.summaryTextView.text = "Flow Starting..."
    }
    return result
  }
  
  func commonDidFinishBlock(block: FlowDidFinishBlock? = nil) -> FlowDidFinishBlock {
    let result: FlowDidFinishBlock = {
      flow in
      block?(flow)
      self.summaryTextView.text = flow.generateSummary()
    }
    return result
  }

Logging

This my most favourite feature when making Flow. It's always hard for developer to trace the console log as there are too many unwanted logs in the console. Even worse in a serious of async operations. Flow will capture all the logs you sent within the opertions and generate a summary for you at the end. Call flow.generateSummary() in the finish block.

For example:

====== Flow Summary ======
4:17:18 PM [DataBucket] start with:
password: 123456
email: [email protected]
4:17:18 PM [MockAsyncLoginOp] WillStart
4:17:19 PM [MockAsyncLoginOp] simulation of login callback
4:17:19 PM [DataBucket] add for accessToken: sdaftadagasg
4:17:19 PM [MockAsyncLoginOp] DidFinish: successfully
4:17:19 PM [MockAsyncLoadProfileOp] WillStart
4:17:20 PM [MockAsyncLoadProfileOp] simulation of success load profile callback
4:17:20 PM [DataBucket] add for profileRefreshDate: 2017-07-13 08:17:20 +0000
4:17:20 PM [MockAsyncLoadProfileOp] DidFinish: successfully
4:17:20 PM [DataBucket] end with:
profileRefreshDate: 2017-07-13 08:17:20 +0000
email: [email protected]
accessToken: sdaftadagasg
password: 123456
Flow isSuccess: true
======    Ending    ======

====== Flow Summary ======
4:17:06 PM [DataBucket] start with:
password: 123456
email_address: [email protected]
4:17:06 PM [MockAsyncLoginOp] WillStart
4:17:06 PM [MockAsyncLoginOp] DidFinish: withInsufficientInputData: can't find <email> in data bucket
4:17:06 PM [DataBucket] end with:
password: 123456
email_address: [email protected]
Flow isSuccess: false
======    Ending    ======

You can trace the data changes, how the operations run in one place. You can send the summary string to your server if needed.

Why not RxSwift?

Surely RXsSwift is much more powerful in some aspects. BUT I think it's always good if we can make our code: Simple and Human readable With Flow, even code reviewers and non-programmer can understand your logic in the blueprint.

Requirements

Swift 3.2 iOS 8.0

Installation

Flow is available through CocoaPods:

pod "Flow-iOS"

Import:

import Flow_iOS

Author

Roy Ng, [email protected] @ Redso, https://www.redso.com.hk/

Linkedin: https://www.linkedin.com/in/roy-ng-19427735/

License

Flow is available under the MIT license. See the LICENSE file for more info.